9
\$\begingroup\$

What is a good decent power supply noise?

Let me expand, two cases, I have a bench top PSU, I put my scope in AC coupling and look at the ripple it is around 20mV. Is this a good number for a decent PSU? (I am fiddling with ANalog circuits so 20mV noise is big deal)

Second case is my on board regulator, I have a booster that takes 2V to 5V. I look at the 5V without any load and I see a 7mV ripple (saw). Is this normal? I have all the decoupling caps there, so I would have expected much less especially without a decent load.

Bonus question, what is the best way of measuring power supply noise? I guess especially with small currents such as this, there has to be more than touching with a probe?

\$\endgroup\$
  • \$\begingroup\$ How your oscilloscope is powered ? is it on the same ground/mains strip ? Did you try to measure with battery powered, floating oscilloscope ? \$\endgroup\$ – user924 Jul 8 '11 at 13:15
  • 2
    \$\begingroup\$ I am asking, because the scope can give you noisy reading if probe ground is too long etc. Bob Pease has nice article about scope reading national.com/rap/Story/0,1562,18,00.html And first thing I would try is to power off the supply (dont diconnect anything) and see if scope still shows 20 mV ripple. \$\endgroup\$ – user924 Jul 8 '11 at 13:21
  • 1
    \$\begingroup\$ Two points I haven't seen mentioned here are: - Measure the voltage ripple under the expected load conditions (for example if you expect to draw 100 mA, measure the ripple at that current draw from the power supply/regulator) - Measure with a scope that is set on 1M\$\Omega\$ impedance, not 50 \$\Omega\$, as it will distort your results less. \$\endgroup\$ – Joel B Jul 8 '11 at 16:35
  • 1
    \$\begingroup\$ @Rocket Surgeon, totally OT but RIP Bob Pease: en.wikipedia.org/wiki/Bob_Pease#Death I hope his stuff lives on the internet forever. \$\endgroup\$ – Mark Ransom Jul 8 '11 at 22:22
13
\$\begingroup\$

There is of course no single answer to what "decent" power supply noise is. That's like asking what a decent car is without telling us whether its for driving around a racetrack or back country dirt roads.

Whether the values you mention are decent depends on how that power rail will be used. What you really seem to be asking is just from the power supply's point of view whether these value seem reasonable or not. 20mV for a generic bench top supply sounds quite reasonable to me, and so does 7mV for a on-board boost converter (in fact that's actually quite good compared to a lot of them).

Your circuit, however, may have a different opinion. If the 5V supply is just powering digital circuitry, then it's a lot cleaner than it needs to be. Even 100mVpp ripple would be tolerable.

If you're powering sensitive analog circuitry, then 7mV could be large. In that case the frequency content of the ripple also matters. Most analog ICs have a power supply rejection spec. There is active electronics in the IC to make its operation somewhat independent of the power supply voltage. However, that electronics can only react to noise up to some frequency. The frequency requirements to get the specified power supply rejection ratio is rarely specified. It's a good practice to put a ferrite bead or small chip inductor followed by a ceramic cap to ground on the power leads of analog parts. This will attenuate the high frequencies of the noise, with the remaining low frequencies hopefully in the range the part can handle and reject actively.

Some parts are much more susceptible to this than others. The first time I used one of the Freescale multi-axis accellerometers there was a lot of noise on the output. The power supply noise actually seemed to be amplified onto the output. Adding the aforementioned chip inductor in series with cap to ground on the power lead helped a lot to clean up the output signal.

To answer your last question, the normal way to look at power supply noise is exactly what you did. AC couple the scope input, crank up the gain, and look at the size of the resulting mess.

\$\endgroup\$
  • \$\begingroup\$ Then there are those that power "digital" circuits like ADCs with a 100mVpp noisy power supply and think they can pull 16 bits of precision and assume any noise is on the line. I have also had issues with accelerometers before. Had to mount it to foam before a large amount of the noise was removed. \$\endgroup\$ – Kortuk Jul 8 '11 at 14:06
  • \$\begingroup\$ It would be interesting if someone spent time discussing white noise and such, as if it is 100mVpp of one specific frequency filtering is easy, if it is White noise, well that is a different beast. \$\endgroup\$ – Kortuk Jul 8 '11 at 14:07
  • \$\begingroup\$ @Kortuk: The point of the filtering I was mentioning was not to get rid of all noise, but to get rid of the high frequencies so that what's left over can be dealt with by the IC's power supply rejection capability. So white noise wouldn't make much difference. The filter takes out the high frequencies and the active electronics deals with the rest. \$\endgroup\$ – Olin Lathrop Jul 8 '11 at 21:10
  • \$\begingroup\$ you already received your +1 from me. I try to use comments to add extra information to an already great answer. I thought it would be interesting if you were to explain in some detail how to tell the different types of noise and what they may imply. Or just explain if the 100mVpp is one frequency that is a sign of a problem. \$\endgroup\$ – Kortuk Jul 8 '11 at 23:26
11
\$\begingroup\$

I designed an extremely low-power PSU before, so let me share a graph I made for a presentation where I outlined the difference in noiselevels of various PSUs. The graph shows the logarithmic noiselevel as a function of frequency from DC to 50 kHz. I don't remember how the scale on the Y-axis is offset but you can get the general gist of it from the description:

  • Red curve: representing a typical digital product's 3.3v supply (under use), it was in your 10 mV range of noise I remember
  • Purple curve: typical wall-wart plus a low-noise 5.6V LDO
  • Blue curve: the above plus another 5V regulator
  • Black curve: my PSU design, which had around 1-3 uV of noise

So depending on the amount of filtering and design you do, the PSU noise can differ by 4 orders of magnitude! your 20 mV from a bench-top PSU is pretty good and standard I think (see caveat below for oscilloscope probe noise).

Normal oscilloscopes are pretty much worthless for any work below 10 mV, by the way. You also want to look at the fourier-transform (spectral content) of the noise to draw any useful conclusions. Of course if you see something simple like a major ripple or instability this is a good start but often noise is not that obvious.

Dedicated spectrum-analyzers are the way to go, but normally they are for RF use and go from something like 100 kHz to 5 GHz - not very interesting if you are debugging an analogue audio amplifier for example. Some of the older models go from DC to say 100 kHz.

You also need to couple the measurement point to the instrument by something else than a (normal) oscilloscope probe. You easily add dozens of mV of noise just by the ground-loop from the probe. Probes with an integrated earth-lead can be used but best is a dedicated coaxial connector and cable from your PCB.

Logarithmic noiselevels of various PSUs

\$\endgroup\$
3
\$\begingroup\$

Most switching power supplies that I've been involved with the design on specify 1% of the rated DC output as maximum peak-to-peak ripple; 50mV for a 5V rail, 120mV for a 12V rail, etc.

Linear supplies tend to be much less noisy, as there's no HF switching ripple component on the output.

It's not uncommon for a switching power supply rail to have multiple LC filter stages, or feed a linear regulator stage if extra-low ripple is needed.

Ripple measurement is an art form in itself. You have to take measures not to pick up common-mode noise. Often, the oscilloscope used for the measure is set to reduced bandwidth (20 MHz is common) and capacitors are used to get rid of 'extraneous' HF (keeping the switching ripple and line frequency components visible) - 100nF in parallel with 10uF is not unheard of. Sometimes, a \$50\Omega\$ resistor is used as a load (along with the capacitors) and the connection to the scope is made with a shielded coaxial cable.

\$\endgroup\$
1
\$\begingroup\$

Those look like normal noise levels on a power supply line, but this doesn't mean that you have that much noise on your analog signal. The Power Supply Rejection Ratio PSRR is the factor that describes how much of the power supply noise is superimposed on the signal, for example in an opamp data sheet.

\$\endgroup\$
1
\$\begingroup\$

Datasheets for the two benchtop PSUs I use specify 15-30 mVpp voltage ripple in the range 20 Hz - 20 MHz.

Everything above 100 kHz-1 MHz is cut off by decaps.

To cut off below 100 kHz:
1) a linear on-chip regulator
2) a ferrite choke (together with capacitors to ground) between the power source and the power consumer
can be used.

When I first realized there is such a "large" fluctuation of power supply (around 10-20 mV), I was scared. However, after putting the transient noise in my CAD 100 kHz noise was a nearly flat line (I usually do simulations for units of microseconds, while T=1/100 kHz=10 us). This is because often digital and analogue electronic devices work with Mega and Giga Hz frequencies.

But it depends on the application and working frequency of a device under test.

PS: to say for sure whether it impacts your device or not, put the VDD transient noise to your simulator and see whether it affects results or not.

\$\endgroup\$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.